RAS Dataset

The current challenge in effectively treating atrial fibrillation (AF) stems from a limited understanding of the intricate structure of the human atria. The objective and quantitative interpretation of the right atrium (RA) in late gadolinium-enhanced magnetic resonance imaging (LGE-MRI) scans relies heavily on its precise segmentation. Leveraging the potential of artificial intelligence (AI) - based approaches for RA segmentation presents a promising solution. However, the successful implementation of AI in this context necessitates access to a substantial volume of annotated LGE-MRI images for model training. In this paper, we present a comprehensive 3D cardiac dataset comprising 50 high-resolution LGE-MRI scans, each meticulously annotated at the pixel level. The annotation process underwent rigorous standardization through crowdsourcing among a panel of medical experts, ensuring the accuracy and consistency of the annotations. Our dataset represents a significant contribution to the field, providing a valuable resource for advancing AI-based RA segmentation methods.

当前有效治疗心房颤动（atrial fibrillation, AF）所面临的核心挑战，在于对人类心房复杂解剖结构的认知不足。对钆延迟增强磁共振成像（late gadolinium-enhanced magnetic resonance imaging, LGE-MRI）扫描中的右心房（right atrium, RA）开展客观定量解读，高度依赖其精准的图像分割。借助基于人工智能（artificial intelligence, AI）的方法实现右心房分割是颇具前景的解决方案，但在此场景下成功落地人工智能技术，需获取大量带标注的LGE-MRI图像用于模型训练。本文提出一套完整的3D心脏数据集，包含50例高分辨率LGE-MRI扫描图像，每例均经过像素级精细标注。标注流程由医学专家专家组开展众包，并经过严格标准化处理，确保了标注结果的准确性与一致性。本数据集为该领域提供了宝贵的研究资源，有助于推动基于AI的右心房分割方法的发展，是该领域的一项重要贡献。